Seed1.5-VL

Seed1.5-VL is ByteDance Seed Team's vision-language foundation model, released with its technical report on May 12, 2025. It pairs a 532M-parameter vision encoder with a Mixture-of-Experts language model with 20B active parameters, and despite the compact architecture it reports state-of-the-art results on 38 out of 60 public VLM benchmarks.

Its strengths skew toward agentic and reasoning-heavy visual work: GUI control, video understanding, visual puzzles, and gameplay, where ByteDance reports it outperforming OpenAI's CUA and Claude 3.7 Sonnet. It is deployed on Volcano Engine under the model ID doubao-1-5-thinking-vision-pro-250428, with a usage cookbook on GitHub.

The honest caveats: the benchmark claims are vendor-reported, and while the cookbook repository is Apache-2.0, the model weights themselves have not been published, so this is an API-only entry.